随着深度学习推理技术的不断发展，让小型设备运行深度学习成为可能，阿里体育等IT大厂，推出的“乐动力”、“天天跳绳”AI运动APP，让云上运动会、线上运动会、健身打卡、AI体育指导等概念空前火热。那么，能否将这些在APP成功应用的场景搬上微信小程序，分享这些概念的红利呢？本系列文章就带您一步一步从零开始开发一个AI运动小程序，本系列文章将使用“AI运动识别”小程序插件，插件详情可以前往微信服务市场搜索相应插件。
一、初始化人体识别功能
人体识别能力调用前需要初始化（可多次初始化），插件的具有双人体识别引擎，默认不指定引擎将自动选择最优引擎，也可指定引擎，识别引擎的差别请参考集成文档。
代码语言：javascript复制const that = this;
humanDetection.initialize({
//ve: 2,
callback(err) {
uni.hideLoading();
if (!err) {
console.log('人体检测能力初始化成功。', humanDetection.getVe());
return;
}
uni.showModal({
content: `初始化失败，详细信息：${err.message}`,
showCancel: false
});
}
});二、调用人体识别功能
代码语言：javascript复制//接上篇抽帧
const context = wx.createCameraContext();
const listener = context.onCameraFrame((frame) => {
//frame.data图像数组将用于后续的人体识别
const image = {
width: Number(frame.width),
height: Number(frame.height),
data: new Uint8Array(frame.data)
};
humanDetection.detectionAsync(image).then(human=>{
console.log(human);
});
});
listener.start();三、人体识别结果
代码语言：javascript复制{
"score": 0.7251838252824896,
"width": 480,
"height": 640,
"timestamp": 1684325583506,
"keypoints": [
{
"y": 103.56424502483996,
"x": 256.0109768927012,
"score": 0.4570312798023224,
"name": "nose"
},
{
"y": 90.71574411931093,
"x": 270.50707531981544,
"score": 0.58251953125,
"name": "left_eye"
},
{
"y": 90.09307140499885,
"x": 244.60145098668588,
"score": 0.552734375,
"name": "right_eye"
},
{
"y": 95.37442872749463,
"x": 286.88540601464354,
"score": 0.75244140625,
"name": "left_ear"
},
{
"y": 93.90985185987653,
"x": 231.85045311881774,
"score": 0.7392578125,
"name": "right_ear"
},
{
"y": 162.78484030178413,
"x": 311.0556351088547,
"score": 0.919921875,
"name": "left_shoulder"
},
{
"y": 161.2381417518693,
"x": 203.38841526521654,
"score": 0.77197265625,
"name": "right_shoulder"
},
{
"y": 240.78316880186404,
"x": 310.24842737472574,
"score": 0.84912109375,
"name": "left_elbow"
},
{
"y": 242.12329664294745,
"x": 201.77353663841666,
"score": 0.833984375,
"name": "right_elbow"
},
{
"y": 281.9979693520591,
"x": 352.673287407275,
"score": 0.59765625,
"name": "left_wrist"
},
{
"y": 286.2878520237733,
"x": 173.8384814716242,
"score": 0.63427734375,
"name": "right_wrist"
},
{
"y": 307.2371714929637,
"x": 273.4654390815558,
"score": 0.82177734375,
"name": "left_hip"
},
{
"y": 305.7596342955926,
"x": 222.0470940485152,
"score": 0.84912109375,
"name": "right_hip"
},
{
"y": 417.72422545441424,
"x": 267.0201893540119,
"score": 0.6103515625,
"name": "left_knee"
},
{
"y": 414.7774591985668,
"x": 204.46947287192143,
"score": 0.8701171875,
"name": "right_knee"
},
{
"y": 481.8868752974563,
"x": 280.1792094531294,
"score": 0.74072265625,
"name": "left_ankle"
},
{
"y": 467.38751675509405,
"x": 247.29222232381437,
"score": 0.7451171875,
"name": "right_ankle"
}
],
"bodyRange": {
"x": 173.8384814716242,
"y": 90.09307140499885,
"width": 178.8348059356508,
"height": 391.79380389245745
}
}四、识别结果旋转矫正
获得人体识别结果后，可以调用rotate()进行整体旋转，也可以调用rotateToUpright()自动旋转直立图，进行横、竖屏适配。
下一篇将为您介绍人体骨骼图绘制...
